[Eeglablist] ICA clustering
Dr. Peter Ullsperger
PUllsperger at t-online.de
Wed May 24 00:43:00 PDT 2006
I have been clustering ICA components of N400 activity and the same questions as you asked emerged:
Basically, the cluster analysis as a statistical approach cannot answer our questions we want to solve from the neurophysiological point of view, i.e. clustering may simply help to put togehther some ICs with related properties and to sort out others. (This is particularly important if one has the problem to overview data of some 60 ICs and 20 subjects.) Therefore, we have duly to consider hypotheses and previous knowledge! (It´s not a method for "blind" solutions).
There is no recipe for weighting and number of preclustering parameters and my impression is that one has to play with to get final clusters which can be physiologically interpreted. Also the possibility of computing subclusters may be considered. (There are also other methods of clustering - not implemented in EEGLAB - which determine the number of clusters "automatically").
With the k-mean clustering one has to define the number of clusters beforehand: My approach is to put such a number that a cluster can obtain at least one IC of each subject.
But as you realized, mostly you get clusters which contain more than one IC of a subject and in addition some subjects are totally missing in this very cluster. Now you have the possibility to edit this cluster, but the question whether one cluster may contain several ICs of one and the same subject cannot be solved by editing. One should consider whether it might be reasonable that several independent activities (similar in some and different in some other
properties) may be important for and contribute to the investigated process. In some subjects this activity might have for instance another dipole localization preventing to be included into a certain cluster (i.e., also the weighting of IC properties should be a matter of hypothesis and/or previous knowledge).
Regarding the handscreening of EMG artifacts, in my analyses the nearly perfect inclusion of muscle artifact Ics in one cluster and the eye blink ICs in another one I considered as an indication that my selected weightings were rather adequate.
Sorry, no recipe but challenge!
> We've been working with attentional blink task data (64 channel) for 20
> subjects using EEGLAB (version 5.02) and are currently attempting to
> cluster ICA components. There appears to be a wide range of results
> produced depending on the chosen clustering parameters. Some experienced
> advice on the following would be greatly appreciated:
> weight and number of pre-clustering parameters
> reasonable PCA dimensions for ERP, spectrum, ERSP/ITC
> better to reduce the number of parameters used for pre-clustering than use
> final dimension reduction
> number of clusters to compute (Should this be based on ERP literature? Is
> it reasonable to have a cluster with several components per subject?)
> when using dipoles it appears to be necessary to rather carefully hand
> screen artifact components (EMG components are sometimes modelled by
> dipoles with RV < 10% and "good" components sometimes have relatively large
> RVs ~30%) How much hand screening is necessary for good results?
> Thanks in advance,
> Larry Greischar
> Heleen Slagter
> University of Wisconsin
> eeglablist mailing list eeglablist at sccn.ucsd.edu
> Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
> To unsubscribe, send an empty email to eeglablist-unsubscribe at sccn.ucsd.edu
More information about the eeglablist